Predicting a conversion rate
Aspects of the present disclosure involve a system comprising a storage medium storing a program and method for predicting a conversion rate. The program and method provide for receiving, from an advertisement service, a bid to display a first advertisement at a computing device; determining, in response to receiving the bid, a set of features that relate to the first advertisement; providing the set of features to a machine learning model configured to output a predicted conversion rate for the first advertisement, the machine learning model having been trained based on multi-task learning using plural sets of features corresponding to plural second advertisements, the plural sets of features being associated with both click-through conversions and view-through conversions; and determining, based on the output of the machine learning model with respect to the set of features, the predicted conversion rate for the first advertisement.
1 . A system comprising:
at least one processor;
at least one memory component storing instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
receiving, from an advertisement service, a bid to display a first advertisement at a computing device;
determining, in response to receiving the bid, a set of features that relate to the first advertisement;
providing the set of features to a machine learning model configured to output a predicted conversion rate for the first advertisement, the machine learning model having been trained based on multi-task learning using plural sets of features corresponding to plural second advertisements, the plural sets of features being associated with both click-through conversions and view-through conversions;
determining, based on the providing, model loss by performing inverse propensity weighting with respect to an auxiliary task that is not used to train a click-through-rate model; and
determining the predicted conversion rate for the first advertisement, the predicted conversion rate being based at least in part on the model loss,
wherein the model loss is calculated as follows:
sum(loss)=loss( ctc= 1)*1/ p _swipe+loss( vtc )*1+loss(negative)*1,
with ctc representing click-through conversions, vtc representing view-through conversions, p_swipe representing a propensity score for the auxiliary task, and negative representing negative transfer to different tasks.
2 . The system of claim 1 , wherein determining the predicted conversion rate is further based on performing inverse propensity weighting with respect to the plural sets of features corresponding to the plural second advertisements.
3 . The system of claim 2 , wherein the predicted conversion rate is debiased at least in part by the inverse propensity weighting in conjunction with the multi-task learning.
4 . The system of claim 1 , wherein the multi-task learning is associated with tasks for pixel events, the pixel events comprising pixel page view, pixel sign up, pixel add to cart and pixel purchase.
5 . The system of claim 1 , the operations further comprising:
embedding a subset of the set of features that relate to the first advertisement; and
concatenating, based on the embedding, the set of features for providing as input to the machine learning model.
6 . The system of claim 5 , the operations further comprising:
providing the concatenated set of features to a deep and cross network, the deep and cross network comprising multiple cross layers configured to model explicit feature interactions, the deep and cross network further comprising a deep network configured to model implicit feature interactions,
wherein output of the deep and cross network is provided as input to the machine learning model.
7 . The system of claim 1 , wherein the plural sets of features are further associated with advertisement impressions for a preset time period, swiped advertisement identifiers for the preset time period and other advertisement identifiers.
8 . The system of claim 1 , wherein the multi-task learning corresponds to a progressive layered extraction (PLE) model.
9 . The system of claim 1 , wherein the predicted conversion rate corresponds to a post-click conversion rate.
10 . A method comprising:
receiving, from an advertisement service, a bid to display a first advertisement at a computing device;
determining, in response to receiving the bid, a set of features that relate to the first advertisement;
providing the set of features to a machine learning model configured to output a predicted conversion rate for the first advertisement, the machine learning model having been trained based on multi-task learning using plural sets of features corresponding to plural second advertisements, the plural sets of features being associated with both click-through conversions and view-through conversions;
determining, based on the providing, model loss by performing inverse propensity weighting with respect to an auxiliary task that is not used to train a click-through-rate model; and
determining the predicted conversion rate for the first advertisement, the predicted conversion rate being based at least in part on the model loss,
wherein the model loss is calculated as follows:
sum(loss)=loss( ctc= 1)*1/ p _swipe+loss( vtc )*1+loss(negative)*1,
with ctc representing click-through conversions, vtc representing view-through conversions, p_swipe representing a propensity score for the auxiliary task, and negative representing negative transfer to different tasks.
11 . The method of claim 10 , wherein determining the predicted conversion rate is further based on performing inverse propensity weighting with respect to the plural sets of features corresponding to the plural second advertisements.
12 . The method of claim 11 , wherein the predicted conversion rate is debiased at least in part by the inverse propensity weighting in conjunction with the multi-task learning.
13 . The method of claim 10 , wherein the multi-task learning is associated with tasks for pixel events, the pixel events comprising pixel page view, pixel sign up, pixel add to cart and pixel purchase.
14 . The method of claim 10 , further comprising:
embedding a subset of the set of features that relate to the first advertisement; and
concatenating, based on the embedding, the set of features for providing as input to the machine learning model.
15 . The method of claim 14 , further comprising:
providing the concatenated set of features to a deep and cross network, the deep and cross network comprising multiple cross layers configured to model explicit feature interactions, the deep and cross network further comprising a deep network configured to model implicit feature interactions,
wherein output of the deep and cross network is provided as input to the machine learning model.
16 . The method of claim 10 , wherein the plural sets of features are further associated with advertisement impressions for a preset time period, swiped advertisement identifiers for the preset time period and other advertisement identifiers.
17 . The method of claim 10 , wherein the multi-task learning corresponds to a progressive layered extraction (PLE) model.
18 . The method of claim 10 , wherein the predicted conversion rate corresponds to a post-click conversion rate.
19 . A non-transitory computer-readable storage medium storing instructions that, when executed by at least one processor, cause the at least one processor to perform operations comprising:
receiving, from an advertisement service, a bid to display a first advertisement at a computing device;
determining, in response to receiving the bid, a set of features that relate to the first advertisement;
providing the set of features to a machine learning model configured to output a predicted conversion rate for the first advertisement, the machine learning model having been trained based on multi-task learning using plural sets of features corresponding to plural second advertisements, the plural sets of features being associated with both click-through conversions and view-through conversions;
determining, based on the providing, model loss by performing inverse propensity weighting with respect to an auxiliary task that is not used to train a click-through-rate model; and
determining the predicted conversion rate for the first advertisement, the predicted conversion rate being based at least in part on the model loss,
wherein the model loss is calculated as follows:
sum(loss)=loss( ctc= 1)*1/ p _swipe+loss( vtc )*1+loss(negative)*1,
with ctc representing click-through conversions, vtc representing view-through conversions, p_swipe representing a propensity score for the auxiliary task, and negative representing negative transfer to different tasks.
20 . The non-transitory computer-readable storage medium of claim 19 , wherein determining the predicted conversion rate is further based on performing inverse propensity weighting with respect to the plural sets of features corresponding to the plural second advertisements.